Including nearly 20,000 participants, the largest study on amyloid prevalence to date estimates that a third of cognitively normal people older than 70 have amyloid building up in their brains. Published January 31 in JAMA Neurology and led by Olin Janssen and Willemijn Jansen at Maastricht University in the Netherlands, the meta-analysis compiled amyloid PET and cerebrospinal fluid (CSF) biomarker data from participants in research and memory clinic cohorts that are part of the Amyloid Biomarker Study, an ongoing, international, data-sharing initiative that started in 2013. The new analysis doubles the size of a 2015 meta-analysis published by the same group. Among other findings, it reports that even among cognitively sharp 50-year-olds, 17 percent already have amyloid, and that CSF Aβ42 bested amyloid-PET in detecting amyloid build-up, at least among people without dementia. For those with AD dementia, amyloid-PET outperformed CSF Aβ42.

  • A meta-analysis of nearly 20,000 participants calculated the prevalence of brain amyloid across age and cognitive status.
  • A third of cognitively normal people over 70 have amyloid, and ApoE4 carriers have steeper increases with age.
  • Among people without dementia, CSF Aβ42 picks up 10 percent more cases than does PET.

In 2015, the Amyloid Biomarker Study published two separate meta-analyses of amyloid prevalence. In one, the researchers combined data from 91 studies that measured amyloid in more than 7,500 people without dementia, and in the other, they meta-analyzed data from studies of more than 1,800 people with different types of dementia, including AD (May 2015 news). They found that among people without dementia, amyloid prevalence rose with age, and was higher among ApoE4 carriers. Overall, the study found that whether amyloid status was determined by amyloid-PET or CSF Aβ42, one-quarter of people with normal cognition or subjective cognitive decline had amyloid plaques, as did half of people with mild cognitive impairment, and nearly 90 percent of people with a clinical diagnosis of AD dementia. This percentage decreased at high ages, likely due to the increasing prevalence of non-amyloid pathologies that contribute to dementia in older people.

Amyloid Prevalence. Based on PET data, amyloid prevalence rose with age among people with normal or impaired cognition, but declined with age among people with dementia. [© (2022) American Medical Association. All rights reserved.]

At 19,097 participants, the 2022 study more than doubles the size of the previous one. Jansen and Janssen and colleagues compared amyloid prevalence across age, cognitive status, ApoE genotype, and by biomarker modality (i.e., CSF versus PET). A total of 10,139 participants in 50 cohorts had undergone amyloid-PET scans, while 8,958 participants across 51 cohorts had had CSF Aβ42 measured; only 1,571 underwent both.

When the researchers relied on the cutoffs for amyloid positivity provided by each cohort included in the meta-analysis, they arrived at similar estimates of amyloid prevalence as they had in 2015. Among those without dementia, amyloid cropped up in 24 percent of those with normal cognition, 27 percent of people with subjective cognitive decline, and 51 percent of people with MCI. Findings were similar whether amyloid-PET or CSF Aβ42 was used. However, when the scientists recalculated cutoffs based on the distribution of biomarker measurements observed in each cohort, they found that CSF Aβ42 flagged 10 percent more amyloid-positive cases than did PET, raising the prevalence in cognitively normal people to a third. This difference between PET and CSF may change once newer fluid marker assays become widely adopted.

Amyloid prevalence increased with age among those without dementia. For example, based on the adjusted CSF Aβ42 measurements, 17 percent of cognitively normal people between the ages of 50-54 had evidence of amyloid. By age 70, a third did, and by age 95, more than half did.

“These patterns imply that at least during the early stages of AD and before dementia onset, CSF may be a more sensitive marker of amyloid accumulation than PET,” wrote Christina Young and Elizabeth Mormino of Stanford University in a JAMA Neurology editorial. “This finding is consistent with work showing that discordant cases with CSF-based amyloid positivity and PET-based amyloid negativity are more likely to become amyloid-positive on PET at follow up,” they added. Work from other groups also indicates that CSF Aβ levels drop before plaques can be detected by PET (Aug 2016 conference news). 

The size of this study gave the researchers enough statistical power to compare amyloid prevalence across different ApoE genotypes. E4/E4 carriers started accumulating amyloid at the youngest age, followed by E3/E4, E2/E4, E3/E3, and E2/E3 carriers. Notably, the amyloid prevalence among E3/E4 carriers was 10 percent higher than it was among E2/E4 carriers across all groups without dementia, highlighting a protective effect of the E2 allele that had not been observed in the smaller, 2015 study.

For people with a clinical diagnosis of AD dementia, the trends were different. For one, amyloid-PET consistently detected a slightly higher percentage of amyloid-positive people than did CSF Aβ42, even when the scientists recalculated the cutoffs. As reported in 2015, the prevalence of amyloid dipped slightly with advancing age among those with AD dementia, ranging from 91 percent at age 50 to 81 percent by age 95. However, in the current study, this dip was no longer statistically significant. ApoE4 carriers with dementia were more likely to harbor amyloid pathology than their noncarrier counterparts, with a prevalence of 97 percent for homozygotes, 87 percent for heterozygotes, and 80 percent for noncarriers.

As in 2015, amyloid accumulation was similar between the sexes, regardless of age or cognitive status. Among those without dementia, people with more education tended to have more amyloid, suggestive of cognitive reserve in the face of pathology. In people with AD dementia, educational level correlated with more amyloid, but only up to age 60.

“Overall, the present study is valuable because it reflects a large, coordinated effort across 85 cohorts to estimate the prevalence of abnormal amyloid accumulation, an initial key pathological change of AD,” wrote Mormino and Young. “These prevalence estimates can improve recruitment efficiency for clinical trials that target individuals with biomarker positivity.” However, they noted that at the group level, amyloid alone is insufficient to indicate future clinically meaningful progression.

Young and Mormino bemoaned that the study lacks information about race and ethnicity. “It is alarming that a large-scale effort that examined the data of more than 19,000 individuals and characterized them by PET or CSF measures was not positioned to report the role of race and ethnicity or the proportion of individuals composing various racial and ethnic groups to give a sense of the cohort’s demographic characteristics and generalizability of findings,” they wrote. Gil Rabinovici, University of California, San Francisco, agreed, noting that 94.6 percent of the participants came from North America and Europe. “We must do better as a field in recruiting a more diverse, global, and representative sample of research participants to AD biomarker studies,” he wrote (comment below).

Particularly for people in the younger age groups without dementia, the present study found a significantly higher proportion with amyloid pathology than did a smaller, population-based study by the Mayo Clinic Study of Aging (Roberts et al., 2018). While Jansen et al. report that 17 percent of cognitively normal people in their 50s had amyloid, the MCSA study found that only 3 percent in this age group did. Jansen and colleagues found the prevalence of amyloid was also much higher in people with MCI than was reported by the MCSA scientists. The authors attribute the discrepancy to differences in the study populations. Whereas the MCSA is a population-based survey of Olmsted County, Minnesota, the Jansen study comprised numerous research-based and clinical cohorts—likely to be flush with people concerned about their cognitive health.—Jessica Shugart


  1. This is an important initiative, and we are pleased to be able to contribute from the UCSF ADRC. The prevalence estimates from this large and international dataset will doubtlessly serve as benchmarks for clinical interpretation of amyloid biomarkers, clinical trial design, and, hopefully, implementation of effective therapies.

    With the larger dataset, it is nice to see both replication of findings from the 2015 meta-analysis, as well as to have adequate power to detect more subtle effects (e.g., the protective effect of E2/E4 vs E3/E4 genotype). It’s also nice to see congruence between visual and quantitative approaches to classifying amyloid PET scans—we are finding the same in IDEAS—as visual read is still the standard in clinical practice.

    Though this is an important contribution, the study has limitations that need to be addressed in future work. First, 94.6 percent of all participants are from North America or Europe, and race and ethnicity were not reported in most cohorts. I strongly agree with the editorial by Young and Mormino that we must do better as a field in recruiting a more diverse, global, and representative sample of research participants to AD biomarker studies.

    Second, the CSF analysis is based primarily on the older Ineptest assay. The field is moving rapidly toward more precise, high-sensitivity immunoassays, and toward CSF ratios (Aβ42/40 or p-tau/Aβ42) rather than Aβ42 as a stand-alone measure to define amyloid status. The emphasis on the higher sensitivity of CSF compared to PET needs to be counterbalanced by the higher specificity of PET for clinically meaningful overall AD neuropathology (Lesman-Segev et al., 2021). This is a critical point, especially for interpreting the clinical significance of a positive amyloid biomarker in regard to a patient’s cognitive impairment.


    . Diagnostic Accuracy of Amyloid versus 18 F-Fluorodeoxyglucose Positron Emission Tomography in Autopsy-Confirmed Dementia. Ann Neurol. 2021 Feb;89(2):389-401. Epub 2020 Dec 7 PubMed.

  2. We acknowledge that this is an impressive feat of global collaborative science.

    It's interesting that reanalysis of CSF cutoffs leads to increasing prevalence. To some extent this is similar to situations where we simply change cut points for conditions such as diabetes and hypertension with millions overnight coming into that diagnostic category. Clearly, population cohorts such as CFAS give much higher prevalences of "amyloid" pathology (including Aβ deposition, not all of which is in a true amyloid β-sheet state) at death.

    This correlates relatively poorly with dementia status. Some of the existing in-life biomarkers may lack specificity and fail to predict progression (Dubois et al., 2021). Presumably, as in-life techniques become more sensitive, the prevalence that this gives will continue to rise, so prediction of dementia status and progression may require other approaches unless nearly everyone over 65 is to be labelled as having AD. This is important in thinking about the current debate of how AD is defined. 

    The age effect is interesting, not just in terms of rising prevalence but in the convergence between demented and non-demented individuals. We report this effect pathologically in the NEJM paper (Savva et al., 2009). The clinical and demographic context in which these biomarkers are used are really important to being able to interpret the findings. 

    The paper also talks about prodromal/preclinical AD. With ever-increasing biomarker sensitivity, we would expect such approaches to report increasing prevalences of AD over time until “everyone has it” (Brayne and Calloway, 1988). But, as with changing criteria for conditions such as diabetes and hypertension, clinical context and burden of pathology remain important, as not everyone with AD pathology will have, or develop, dementia as the clinical syndrome. Nor are they necessarily on a trajectory of cognitive decline (an interesting area for further work).

    The issue of preclinical AD, that is, changes that are on a trajectory to develop clinical AD-type dementia, still need be defined and validated in “usual” aging populations rather than in volunteer cohorts. We may have a situation where near 100 percent of over 70s could be labelled as such. As we have remarked in editorials and commentaries in the past, this may be good for pharma companies and their countries’ government exchequer, but not necessarily good for health and well-being of the population, including the specificity and meaning of diagnostic labels applied to individuals. 

    The researchers want to show results for the AD spectrum, but it is likely that the use of clinical/research cohorts with many selection/exclusion criteria will not reveal the true continuum of cognition or biomarker status. By applying filters based on discrete phenotypes, and exclusion of other diagnoses/comorbidities, the results are likely to artificially enhance the separation of amyloid prevalence between those defined as normal, MCI, or AD dementia. Out of the 85 cohorts, only three are described as population cohorts, representing fewer than 500 of the 19,097 participants (2.5 percent). Even with such filters, with increasing age amyloid prevalence becomes unlikely to distinguish between clinical AD dementia and MCI, and those with normal cognition approach overlap. These boundaries would likely be more blurred in unselected cohorts, where increasing dementia with age is due to multiple underlying pathologies, and the ability of any one type to predict/explain dementia reduces. 

    The results by APOE genotype are a reminder to interpret the overall findings in light of the fact that many of the cohorts are enriched for APOE e4 carriers (by design or self-selection), as shown by the cohort-level prevalences of e4 often being higher than expected population frequencies. The paper highlights some of the contrasts with Mayo Clinic Study of Aging estimates, including higher prevalence at younger ages and in the MCI group, which may in part be explained by APOE e4. 

    Given the rising prevalence of amyloid even in the cognitively normal with age, who in these studies represent the cognitively fittest end of the spectrum, we are reminded how problematic biomarker definitions of AD are. They may label a large proportion of the population as preclinical/prodromal AD, even when many will live out their lives without clinical dementia. 

    The paper recognizes its limitation regarding the lack of data on race/ethnicity. This raises a caveat when it comes to clinical trial recruitment, which the study aims to inform, in light of robust criticism of the aducanumab trials on these grounds: "approval was based on trials that were not inclusive of the people who bear a disproportionate burden of the disease" (Manly and Glymour, 2021). 


    . Clinical diagnosis of Alzheimer's disease: recommendations of the International Working Group. Lancet Neurol. 2021 Jun;20(6):484-496. Epub 2021 Apr 29 PubMed.

    . Age, neuropathology, and dementia. N Engl J Med. 2009 May 28;360(22):2302-9. PubMed.

    . Normal ageing, impaired cognitive function, and senile dementia of the Alzheimer's type: a continuum?. Lancet. 1988 Jun 4;1(8597):1265-7. PubMed.

    . What the Aducanumab Approval Reveals About Alzheimer Disease Research. JAMA Neurol. 2021 Nov 1;78(11):1305-1306. PubMed.

  3. I would like to congratulate Willemijn Jansen, Olin Janssen, and the leaders of the Amyloid Biomarker Study for another organizational tour de force. (By way of disclosure, I also thank them for enabling us to provide data from one of the 85 cohorts used in the meta-analysis described in this article.)

    Their participant-level meta-analysis provides prevalence estimates for Aβ positivity, stratified for clinical severity (cognitively unimpaired, mild cognitive impairment, dementia), age, sex, APOE genotype, and geographic region, using PET and CSF biomarker data from an unprecedented number of persons around the world.

    Findings from this study will have an important impact on the field. Among other things, they could help to inform AD treatment and prevention trial design and recruitment, clinical care strategies, and the health-economic impact of AD-modifying and prevention therapies now in development. Equally important, the strategies used to generate their biomarker findings could provide an operational roadmap for other efforts of this kind.

    In 2015, the study leaders reported findings from a meta-analysis of Aβ PET data from 7,583 persons without dementia (Jansen et al., 2015) and 1,359 persons with dementia (Ossenkopele et al., 2015). They have since extended their findings to 19,097 persons with no cognitive impairment, mild cognitive impairment and dementia (MCI), including 10,139 who had PET and 8,895 who had CSF biomarkers of Aβ plaque deposition.

    As noted in the article, the thoughtful editorial by Cristina Young and Beth Mormino, and Gil Rabinovici’s Alzforum comment, the report has limitations, the most glaring of which is the unacceptable paucity of participants from underrepresented ethnic and racial groups. This limitation needs to be better addressed in all research studies and clinical trials.

    When emerging blood-based biomarkers (BBBMs) for Aβ positivity and the diagnosis of AD are further developed, tested, and extended to underrepresented groups, there will be an opportunity to use the strategy deployed by this team to provide information about BBBM measurements of amyloid positivity and the diagnosis of AD in even larger and more representative populations.

    As noted in the article, editorial, and comment, there are also some technical limitations that could have influenced  the reported prevalence estimates. Despite those, participant numbers matter, and the perfect should not be the enemy of the good when it comes to those technical issues. The findings of this meta-analysis are still important for stakeholders in the field.


    . Prevalence of cerebral amyloid pathology in persons without dementia: a meta-analysis. JAMA. 2015 May 19;313(19):1924-38. PubMed.

    . Prevalence of amyloid PET positivity in dementia syndromes: a meta-analysis. JAMA. 2015 May 19;313(19):1939-49. PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Meta-Analyses Deliver Most Definitive Data Yet on Amyloid Prevalence
  2. Refining Models of Amyloid Accumulation in Alzheimer’s Disease

Paper Citations

  1. . Prevalence and Outcomes of Amyloid Positivity Among Persons Without Dementia in a Longitudinal, Population-Based Setting. JAMA Neurol. 2018 Aug 1;75(8):970-979. PubMed.

Further Reading

No Available Further Reading

Primary Papers

  1. . Prevalence Estimates of Amyloid Abnormality Across the Alzheimer Disease Clinical Spectrum. JAMA Neurol. 2022 Mar 1;79(3):228-243. PubMed.